1  Foundations

Causal inference is a way of thinking before it is a set of methods. The goal of this part is to build a reliable causal mindset: how to turn real questions into causal estimands, how to spot bias, and how to communicate assumptions.

By the end of Foundations, you should be able to look at a claim like “\(X\) improves \(Y\)” and ask the right follow-up questions:

1.1 What you’ll be able to do

  • Translate a real-world question into a causal estimand (ATE, ATT, LATE; total vs direct effects)
  • Draw DAGs to make assumptions explicit and to decide what to adjust for (and what not to)
  • Recognize the major bias families: confounding, selection/collider bias, post-treatment bias, and measurement issues
  • Match a question + data structure to an identification strategy (experiments, DiD, RDD, IV, SCM)
  • Write conclusions with appropriate scope and uncertainty (what you learned, what you assumed, what remains fragile)

1.2 How to use this part

Read it once quickly, then revisit it while you work on applications. The most important habit is: design first, estimation second.

When you get stuck, use this loop:

  1. Write the estimand in words (and ideally in potential-outcomes notation)
  2. Draw a DAG of your best current causal story
  3. Identify backdoor paths and selection mechanisms
  4. Choose a design that makes the counterfactual credible
  5. Only then choose an estimator and code

1.3 Chapters in this part

  • Why causality? Why prediction and explanation are different tasks, and why counterfactuals are central.
  • Correlation vs causation. Confounding, colliders, and “bad controls” using DAG intuition.
  • Causality for decisions. Why the goal is often better decisions, not just “significant results”; how costs, benefits, and uncertainty enter.

1.4 A simple running example (we will reuse it)

Throughout the book we’ll return to a few recurring stories—training programs, policy reforms, and interventions in systems—because causal reasoning is easier when you repeatedly test your understanding against concrete cases.

If you’d like, pick one question now and keep it as your running example:

“What is the effect of \(X\) on \(Y\) for population \(P\) over horizon \(T\)?”

We’ll refine this question as you learn new tools.